From: AAAI Technical Report WS-98-13. Compilation copyright © 1998, AAAI (www.aaai.org). All rights reserved. A Sample of KnowledgeManagementProjects at George WashingtonUniversity Carl Moore, Seung Ik Baek, and Jay Liebowitz* Dept. Of ManagementScience George WashingtonUniversity Washington, D.C. 20052 Tel: 202-994-0554 Fax: 202-994-4930 JAYL@GWIS2.CIRC.GWU.EDU *RWD TechnologiesDistinguishedProfessor, Universityof Maryland-Baltimore County,as of Aug.17, 1998 ABSTRACT Artificial intelligence (A_r) has beenlookingfor a "killer applieation" since its ineeption. Now,with the worldwideweb, intranets, intelligent agents, and the emergingknowledgemanagement trends, AI mayhavefound its muchneededkiller application. Throughthe use of web-basedintelligent agent technology, knowledgemanagement systems ean be developedin very effeetive ways. This paper will highlight several of the knowledgemanagement projects that have been developedin the Sehoolof Business and Public Management at GeorgeWashingtonUniversity (GWU) whieh incorporate the use of intelligent agents for makingAI integral part of knowledgemanagement. 1.0 An Intelligent-Agent Based Information Warfare Advisor Developed for the US Army War College (USAWC) The capstone, and perhaps most innovative, activity at the USArmyCollegeis the "Strategic Crisis Exercise (SCE). The SCEis two weekexercise/simulation whichserves as the laboratory for applying the students’ knowledgein strategic leadership,military doctrine, policy, and decision makinggained over the student’s tenure at the WarCollege. The SCEinvolves all the War Collegestudents (320+)role playing certain organizations, (e.g., State Department,Joint Chiefs of Staff, Treasury, Military Commands, etc.) wherebya myriadof activities occur in which the students needto react and formulatedecisions. Typically, over 120 subject matter experts are used throughout the SCEto bring somemeasure of reality to this exercise. Oneof the themesthat is integrated throughoutthe SCEis the influence of information warfare (IW). In the SCE,certain civilian and defense-related IWevents occur, and the students needto consult the "IWsubject matter expert" in order to help resolve someof these issues. To help in this regard, an on-line IWadvisor, using intelligent agent technology,wasbuilt by George 42 WashingtonUniversity for USAWC student use during the SCE.Instead of bringing in many experts fromaroundthe country to participate in the SCE,the use of intelligent agents for on-line advisors maybe a cost-effective and efficient techniquefor allowingthe students to interact, pose questions, and engagein discussions with these "surrogate experts" over the SCEintranet. 1.1 SystemGoals In exploratorydiscussions held at the US ArmyWarCollege with experts in the field of informationwarfare (IW)and its use in exercises held at College, the key goals for the Information Warfare Advisor (IWA)are: Providean on-call advisor in IWfor students at the USAWC taking courses or participating in real-time exercises. This advisor would draw on both public and private knowledge bases about 1Wand IWrelated topics to provide students with requested information on IW. Provide a monitoring advisor whichwould analyze email and other messagetraffic collected during the AWC annual Strategic Crisis Exercise (SCE)and issue a real-time alert of IWactivity or instances detectedin the email or messagethreads. This alert wouldbe part of the SCEsimulations. Provide an assessment advisor whichwould analyze email threads and other message traffic collected during the USAWC annual SCEand developa post exercise set of cases on the lW activity during the SCE.These cases wouldbe used to de-brief students after the exercise, designfuture exercises, as well as add to the IWAknowledgebase on IW. Theinitial research project focusedonly on developinga portion of the IWAas a proof of conceptdemonstrator.Initially, a limited scope advisor agent system was developed, providing the user callable advisor capability for students at the USAWC. The on-call advisor advises students on a limited numberof topics via the user agent, a single brokeragent, and one or two specialist agents in selected IWtopics. The computingenvironmentin whichthe IWAoperates is a heterogeneousnetwork of server and client computers.Theservers are a mix of Sun workstations and WindowsNTservers used as mail, file, andwebservers. Theclients are end-user workstationsconsisting of mostlyIntel 486 and Pentiumclass desktop or laptop PCS. All servers and clients are networkedusing 10 MbpsEthernet and the TCP/IPprotocol. The Relational Database ManagementSystem (RDBMS) installed is the Oracle system. The networkis set-up as a private intranet. Thereis no external accessto the Internet at this time, althoughsuch access is held out as a future potential and mustbe accountedfor in the architecture for expansion. The IWknowledge bases which the 1WA uses are a combinationof existing legacy knowledgebases and databases, as well as new knowledgebases created as part of this research effort and subsequentdevelopmentefforts for the IWA.These knowledgebases and databases are distributed throughout the computing environment,residing on different servers. 1.2 General Approach Thedistributed nature of the knowledge bases in the IWAenvironmentled to the decision 43 to base the IWAconceptual architecture on a distributed AI approachusing a collection of cooperatingintelligent agents. This approach offers both several advantages,as well as certain disadvantages. Howeverit was felt the advantagesof this approach,in lieu of developing a single monolithicexpert system advisor, more than outweighedthe disadvantages. The multi-agent approachis a more complexdesign, and consequently was more difficult to implementthan the monolithic approach. Howeverthe multi-agent approachis better suited to the distributed environment for the IWA.A multi-agent system allows better shared use of the computingresources available on the networkfor load leveling, andavoids the potential bottleneckof a single point of accessfor the advisor. Themulti-agentsystemis also easily extendedby addingagents with different specialties. In this mannerthe capability of the overall IWAsystem can be increased by re-using the developedagent architectures over and over, but each time with a different knowledgebase. 1.3 ConceptualArchitecture--Brokered Agents Thebrokeragents in this architecture act as generalists in their field. Eachbrokermaintains knowledgeof specialist agents in the environment whichhad problemsolution capabilities within the broker’s general domain.Other agents seeking assistance with a problemthen need only to contact the brokers to receive informationon what specific agents wouldbe of assistance, and then contact those agents only, rather than placing a universal query. Optionally, the system could be devised so that the problemcould be divided into larger pieces by the originating agent and passed to the brokers. Thebrokers wouldthen handle the job of further breakingdownthe portion of the problemwithin their owndomain,using domain specific knowledgeabout problemsolving, contact the agents within their brokeragefor solutions, synthesizetheir portions and handoff to the originating agent. Theoriginating agent would then synthesize the partial answersfroma few brokers rather than from a larger numberof specialists. This secondvariant of the broker process, using the brokers as domaingeneralist for sub-problemplanning and solution through subcontractingto specialists is the approachused for the IWA.The broker concept of a cooperative systemof agents is called an agency. The conceptual process of problem solution in a brokeredagencysystembegins with a user presenting a problemto the systemfor solution. Thepresentation of the problemis via a user agent. Theuser agent consists of both a user interface as well as the underlyinglogic and reasoningability to break the probleminto portions and pass those portions to a brokeror brokers for solution. This passing of problem portions is termed subcontracting. Underthe terms of the subcontract the user agent would request assistance, and the broker wouldaccept the problem.In the event no broker was able to solve the problemportion, or no agent was available, then no subcontract wouldbe madeand the user is notified of the failure to perform. If the brokercan solve the problem,then the broker wouldbreak the subcontract into portions for solution by the specialist agents within its domain.Eachspecialist agent could proceedindependently,or in conjunctionwith other agents, to solve the problem.There is no restriction on specialist agents possessingthe sameability as the user agent to subcontractparts of its assignedproblemto other agents througha broker, either the sameor a different broker. Thespecialist agents also have access to both private databases and knowledgebases of their own,as well as public databases and knowledgebases accessible by all specialist agents. These public databases and knowledge bases maybe legacy databases madeavailable to the populationof agents in a standard format wrapperagent. Thejob of the wrapperagent is to act as a translator for the systemfromthe unique format and structure of the legacy databaseto the general structure and format used by the agent system. The brokered agency conceptual architecture has advantagesand disadvantages. Maintenanceand expansion are moredifficult since new agents must be somehow registered with one or morebroker agents so the brokers knowthey exist and whattheir capabilities are. In like manner,newbroker agents must also be made knownto the user agent. This registration can be mademanualat the time of adding or updating agents, increasing the load on maintenance programmers,or can be madeautomatic through logic includedin the designof the agent architecture. However this increases the load on the agent developers,and indirectly on the maintenanceprogrammerssince nowthe agent designs would be more complexand hence more costly to maintain. However,there are a numberof key advantagesoffered by the brokered agency architecture that outweighthe disadvantages. With a brokeragecooperative agreements,or subcontracts, communicationamongthe agents is easier to establish. Thebrokerageconcept reduces sharply the numberof other agents a user agent must be awareof in order to process a problem. Eachbrokeracts as a generalist in its domain, providinga source of informationon the specific agents whichcan be of assistance, or actually taking on the sub-task itself for solution. Brokers can also call uponother brokers, not just the specialist agents within its domain,thereby wideningits field of expertise, as well as providing the system with a hierarchy of domain knowledge. Somebrokers would have more general knowledgethan others, each level of broker breaking the probleminto sub-tasks for subcontractingto other morespecific brokers or to specialist agents. This structure providesthe systemwith a large degree of modularity, making expansioneasier to plan, execute, test, and manage. Theadvantagesoffered by the brokered agencyconceptweresufficient to warrant its adoptionas the conceptualarchitecture of choice for the IWAproject. 2.0 A Web-Based, Intelligent Agent Approach for Managing Knowledge Relating to Storyboard Development for Multimedia Designers in a Virtual Setting Another GWUknowledge management project that used AI/intelligent agent technology involved web-basedaccess for jointly developing a storyboardfor a multimediaeffort. In this context, knowledgemanagement is a methodfor systematically and actively managingand 44 leveraging ideas and design decisions amongteam memberswhile developing storyboards. This research involved developing a knowledge managementsystem to help multimedia team designers create, exchange,and share their storyboards(the storyboardis the script/thread/flowof activities relating to a multimediaproject). Therewerethree intelligent agents usedin this knowledgemanagement architecture. These agents were created using Javascript, HTML, and CFML(Cold Fusion MarkupLanguage). The knowledgecreating activity was handledby the user agent whichremembered all user activities, and dynamicallyorganized a person’s agenda. The knowledgesecuring activity was handled by the knowledgeagent to index knowledge,detect inconsistencies and generate recommendations, and save/retrieve/update knowledge.The knowledge distributing and retrieving activities were performed by the knowledgemanagerto monitorall changesthat occurred in a knowledge repository and forwardthemto the user agent, reformulate queries based on an ontology, and dynamicallyretrieve annotations and generate hyperlinks for them. Thesethree agents have well-defined goals and knowledgeto performtheir tasks (goaloriented). Theyperformtheir tasks without direct humanintervention (autonomousand selfstarting). Theagents interact with each other to performtheir tasks (cooperative). The agents interact with a user database, annotationdatabase, and storyboard database. Throughthis architecture, the multimedia designers used in the research studies could effectively and efficiently developtheir storyboardsin a collaborative fashion via the web in a virtual setting. 3.0 Summary In order for knowledgemanagement to succeed,AI technologies(like intelligent agents) are critical to the mainknowledgemanagement functionsof creating, securing, capturing, combining,distributing, and retrieving knowledge. Thetwo projects briefly describedin this paper highlight someof the workin whichAI can 45 greatly contribute to knowledgemanagement systems and activities. At GWU, weare furthering our workin this area by also examining performanceand learning amongintelligent agents for developing improved knowledgemanagement systems. Lastly, weare looking at developinga methodology for valuating intellectual capital (particularly, humancapital) in an organization. Selected References Liebowitz, J. (ed.), The Handbook of Applied Expert Systems, CRCPress, Boca Raton, 1998. Liebowitz, J. and T. Beckman,Knowledge Organizations: WhatEvery ManagerNeeds to Know,St. Lucie Press/CRCPress, Boca Raton, 1998. Liebowitz, J. and L. Wilcox(eds.), Knowledge Management and Its Integrative Elements, CRC Press, Boca Raton, 1997.